Hacked: Google confirms Gemini AI attacks other company systems while testing
Last Updated: 2026-09-21 12:16:13
It feels like each AI model is trying to better what the other is doing. First we saw ChatGPT and Claude access systems outside of its purview. And now, Google admits that even Gemini AI hacked into systems of not one but three companies which raises further concerns around how quickly these AI models are evolving.
Show AI Summary
Summarized by AI.
Google confirms Gemini AI hacked other systems.- Gemini breached 3 companies during testing.
- AI guessed credentials & used public info.
According to reports, Gemini got through to the internet and bypassed the local borders and hacked other companies while it was being tested for its own cybersecurity abilities. This is the first time when Google’s AI model has bypassed its commands and breached into third-party systems on its own.
Latest and Popular Mobiles
Google explains how Gemini AI went rogue
Google has shared the details about this adventure by its AI model and it seems to have less concerns for some reasons. The company says the hacking incident happened in May this year, and it was actually done while testing of Gemini was going on with Irregular, a company which specialises in these tests.
Related Articles
During a standard testing evaluation, Gemini found public information online and guessed credentials to access three websites it thought were within the scope of its test, as Heather Adkins, Google’s vice president of security engineering has been quoted saying in a statement about the hacking incident.
The company assures that all the affected companies were informed about the hacks, and the training partner is now making changes so that Gemini doesn’t breach its controlled space again.
One of the three victims of the Gemini AI hacking was made possible as the model made multiple attempts on its own to guess their system passwords. While the other two companies were attacked by using the details found in the public systems.
Google also suggests that Gemini stopped the hacking once it realised what was happening, which might be a small consolation in the whole episode and again raises a red flag over such actions in the first place.
Bigger unanswered questions
After all, what prompted Gemini to access the internet and get personal details to target these companies, and secondly, how did the AI model zero-in on these names to attack. Google doesn’t give out those details but surely needs to look into while internally checking for the logs behind these hacks.